1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
|
#!/usr/bin/env python3
"""
Calculate Fisher's linear discriminant for acoustic models.
This module implements Linear Discriminant Analysis for single
stream Sphinx-III acoustic models.
"""
# Copyright (c) 2006 Carnegie Mellon University
#
# You may copy and modify this freely under the same terms as
# Sphinx-III
__author__ = "David Huggins-Daines <dhdaines@gmail.com>"
__version__ = "$Revision$"
import sys
try:
import numpy
except ImportError:
print("FATAL: Failed to import numpy modules. "
"Check that numpy and scipy are installed")
sys.exit(1)
from cmusphinx import s3lda, s3gaucnt
def makelda(gauden_counts):
"""
Calculate an LDA matrix from a set of mean/full-covariance
counts as output by the 'bw' program from SphinxTrain.
@param gauden_counts: Set of full covariance occupation counts.
@type gauden_counts: cmusphinx.s3gaucnt.S3FullGauCntFile
"""
if not gauden_counts.pass2var:
raise Exception("Please re-run bw with '-2passvar yes'")
mean = numpy.concatenate([x[0] for x in gauden_counts.mean])
var = numpy.concatenate([x[0] for x in gauden_counts.var])
dnom = gauden_counts.dnom.ravel()
# If CMN was used, this should actually be very close to zero
globalmean = mean.sum(0) / dnom.sum()
sw = var.sum(0)
sb = numpy.zeros(var[0].shape, 'd')
for d, m in zip(dnom, mean):
diff = m / d - globalmean
sb += d * numpy.outer(diff, diff)
print("Sw:\n", sw)
print("Sb:\n", sb)
BinvA = numpy.dot(numpy.linalg.inv(sw), sb)
u, v = numpy.linalg.eig(BinvA)
top = list(u.argsort())
top.reverse()
u = u.take(top)
# Remember, the eigenvalues are in the columns, but Sphinx expects
# them to be in the rows.
v = v.T.take(top, 0)
print("u:\n", u)
print("v:\n", v)
return v
if __name__ == '__main__':
if len(sys.argv) < 3:
sys.stderr.write("Usage: %s OUTFILE ACCUMDIRS...\n" % (sys.argv[0]))
sys.exit(1)
ldafn = sys.argv[1]
accumdirs = sys.argv[2:]
gauden = s3gaucnt.accumdirs_full(accumdirs)
lda = makelda(gauden)
s3lda.open(ldafn, 'w').writeall(lda[numpy.newaxis, :])
|
